L4 Data Analysis Supplemental Methods
L4 shotgun sequencing datasets of DNA for eight samples and mRNA for seven samples
were downloaded from the MG-RAST website
(http://metagenomics.anl.gov/metagenomics.cgi?page=MetagenomeProject&project=109).
The pre-filtered datasets ranged from 267,833 to 631,084 reads for the DNA samples and
47,406 to 80,615 reads for the RNA samples. Datasets were pre-filtered for quality and to
remove rRNA sequences from the metatranscriptomic datasets. We used ShotMAP to
functionally annotate the datasets using three databases: the FIGfams database from the
Fellowship for the Interpretation of Genomes (Release 12, 2009.
http://www.nmpdr.org/FIG/wiki/view.cgi/FIG/FigFam) that Gilbert et al. used, as well as
the KEGG Orthology (KO, downloaded June 25, 2013) and the Sifting Families (SFams v 1.0)
databases. The FIGfam database consists of 1,422,349 proteins comprising 213,098
families, the KOs are comprised of 4,078,985 proteins across 18,416 families, and SFams
have 6,098,117 proteins distributed in 345,641 families. To facilitate comparisons across
samples, the eight metagenomic datasets were rarefied to 267,833 reads (the size of the
smallest metagenomic dataset) and the seven metatranscriptomic datasets were rarefied to
47,406 reads (the size of the smallest metatranscriptomic dataset). Reads were translated
in all six reading frames and split on stop codons, and then ORF’s having a length of at least
15 amino acids and a RAPSearch bit score of at least 35 were classified in each of the three
databases according to their best protein sequence hit within each database. Family
abundance was then calculated as the number of residues of the translated metagenomic
read that align to the target protein from the family normalized by the length of the average
family member. Raw abundances for the metagenomic datasets were then corrected for
average genome size using MicrobeCensus. For each database, we calculated physiological
richness (the number of families observed in a sample), Good’s coverage (1 – the number of
families with a single hit/number of classified reads for a sample), Shannon entropy, and
the overall classification rate for the 15 samples with respect to each of the three functional
protein family databases (Supplemental Figure 7 A, B, and C).
Challenges associated with the analysis of low-abundance gene families
We observed several noteworthy results regarding low-abundance gene families in the L4
data. First, most of the observed gene families in the L4 metagenomes and
metatranscriptomes have a low absolute abundance (e.g., only 6% of families had over 100
reads in KEGG and only 1% with FIGfams). This long-tailed skew in the gene family rankabundance distribution is common in the analysis of metagenomes and indicates that, for
most families, differences across samples will be highly subject to stochastic effects. As a
result, caution is warranted when interpreting seasonal or diel trends for individual families
or potentially even pathways. Second, the proportion of reads classified into gene families
varies substantially between samples (S12 Figure). Care should thus be taken when
assessing a family’s inter-sample relative abundance distribution, since this measurement is
highly dependent on classification rate: low classification rates artificially inflate relative
abundance estimates for observed gene families. We believe this issue explains why we find
lower functional diversity in January day metagenomes than was originally reported with
the data; this sample has the lowest classification rate of all metagenomes. Finally, gene
families supported by a single read also play an important role at the level of community
diversity. For example, we observed higher richness in January day compared to January
night metatranscriptomes with all three databases, but much of this was due to families
consisting of a single read. Based on our simulations, we hypothesize that this diversity
caused by the large number of low abundance families in daytime samples is largely real
(i.e., low predicted error rates), despite conflicting with previous studies that found
elevated richness in night-time samples (e.g., (1),(2)). ShotMAP provides the computational
tools and summary statistics necessary to explore and address these sample-specific factors
that may confound intersample comparisons of distinct gene families.
References
1.
Poretsky RS, Hewson I, Sun S, Allen AE, Zehr JP, Moran MA. Comparative day/night
metatranscriptomic analysis of microbial communities in the North Pacific
subtropical gyre. Environ Microbiol [Internet]. 2009 Jun [cited 2015 Feb
28];11(6):1358–75. Available from:
http://www.ncbi.nlm.nih.gov/pubmed/19207571
2.
Gilbert JA, Steele JA, Caporaso JG, Steinbrück L, Reeder J, Temperton B, et al. Defining
seasonal marine microbial community dynamics. ISME J [Internet]. 2012 Feb [cited
2014 Mar 21];6(2):298–308. Available from:
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3260500&tool=pmcent
rez&rendertype=abstract